Aerospace Science and Technology 15 (2011) 595605

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Prediction of aerodynamic characteristics of an aircraft model with and without winglet using fuzzy logic technique
Altab Hossain a, , Ataur Rahman b , Jakir Hossen c , A.K.M.P. Iqbal a , M.I. Zahirul d
a

Department of Mechanical Engineering, Faculty of Engineering, University Industry Selangor, 45600, Kuala Selangor, Malaysia Faculty of Engineering, International Islamic University Malaysia, Malaysia Faculty of Engineering, Multimedia University, Malaysia d Department of Mechanical Engineering, Queensland University, Australia
b c

a r t i c l e

i n f o

a b s t r a c t
This paper describes the potentials of an aircraft model without and with winglet attached with NACA wing No. 65-3-218. Based on the longitudinal aerodynamic characteristics analyzing for the aircraft model tested in low subsonic wind tunnel, the lift coecient (C L ) and drag coecient (C D ) were investigated respectively. Wind tunnel test results were obtained for C L and C D versus the angle of attack for three Reynolds numbers Re (1.7 105 , 2.1 105 , and 2.5 105 ) and three congurations (conguration 1: without winglet, conguration 2: winglet at 0 and conguration 3: winglet at 60 ). Compared with conventional technique, fuzzy logic technique is more ecient for the representation, manipulation and utilization. Therefore, the primary purpose of this work was to investigate the relationship between lift coecients and drag coecients with free-stream velocities and angle of attacks, and to illustrate how fuzzy expert system (FES) might play an important role in prediction of aerodynamic characteristics of an aircraft model with the addition of winglet. In this paper, an FES model was developed to predict the lift and drag coecients of the aircraft model with winglet at 60 . The mean relative error of measured and predicted values (from FES model) were 6.52% for lift coecient and 4.74% for drag coecient. For all parameters, the relative error of predicted values was found to be less than the acceptable limits (10%). The goodness of t of prediction (from FES model) values were found as 0.94 for lift coecient and 0.98 for drag coecient which were close to 1.0 as expected. 2010 Elsevier Masson SAS. All rights reserved.

Article history: Received 19 March 2010 Received in revised form 20 November 2010 Accepted 8 December 2010 Available online 15 December 2010 Keywords: Winglet Lift coecient Drag coecient Fuzzy logic

1. Introduction The aerodynamic eciency of an aircraft can be improved through a wingtip device which diffuses the strong vortices produced at the tip and thereby optimize the span wise lift distribution, while maintaining the additional moments on the wing within certain limits. For this purpose one should be able to produce favorable effects of the ow eld using wing tip and reducing the strength of the trailing vortex with the aid of wingtip devices, e.g., winglets, wing tips of complex plan-form, sails, and various modications of the wingtip side edge. The winglet is cambered and twisted so that the rotating vortex ow at the wing tip creates a lift force on the winglet that has a forward component. This forward lift component acts as a negative drag, reducing the total wing drag. Modern interest in winglets spans the last 25 years. Small and nearly vertical ns were installed on a KC-135A and ight was tested in 1979 and 1980 [24,25]. Whitcomb showed that winglets

Corresponding author. Tel.: +6 03 3280 5122x7187; fax: +6 03 3289 6016. E-mail address: altab75@unisel.edu.my (A. Hossain).

could increase an aircrafts range by as much as 7% at cruise speeds. A NASA contract [26] in the 1980s assessed winglets and other drag reduction devices, and they found that wingtip devices (winglet, feathers, sails, etc.) could improve drag due to lift eciency by 10 to 15% if they are designed as an integral part of the wing. Aviation Partners, Inc. has developed two revolutionary design conceptsthe High-Aspect-Ratio Blended Winglet (currently ying on the Gulfstream II and the Boeing Business JetBBJ), and the Spiroid system. It is reported that the blended winglet reduces drag by eliminating the discontinuity between the wing tip and the winglet [5]. A smoothed version is used on the gently upswept winglet of the Boeing 737-400. Boeing Business Jets and Aviation Partners, Inc. have embarked upon a cooperative program to market conventional winglets for retrot to the Boeing 7xx series of jetliners. Flight tests of Blended Winglets on the Boeing Business Jet 737-400 has indicated signicant overall aerodynamic improvement and resulted in a 7% drag reduction. Similarly, the spiroid wing tip produces a reduction in induced drag [7]. Basic aerodynamic analysis, together with ight testing, has indicated that the Spiroid can be 40% to 50% more effective than a Blended Winglet. It is reported that whereas the Blended Winglet reduces cruise

1270-9638/$ see front matter doi:10.1016/j.ast.2010.12.003

2010

Elsevier Masson SAS. All rights reserved.

596

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

Nomenclature

[Kij ] {Fi} {L i }

p R T v c

Re S L CL D CD

Coecient matrix . . . . . . . . . . . . . . . . . . . . . . . dimensionless Load matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N Signal matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Volt Air density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . kg/m3 Absolute pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N/m2 Gas constant of air . . . . . . . . . . . . . . . . . . . . . . . . . . Nm/(kg) K Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K Free stream velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . m/s Chord length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . m Dynamic viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . kg/(m)(s) Reynolds number . . . . . . . . . . . . . . . . . . . . . . . dimensionless Reference area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . m2 Lift force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N Lift coecient . . . . . . . . . . . . . . . . . . . . . . . . . . . dimensionless Drag force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N Drag coecient . . . . . . . . . . . . . . . . . . . . . . . . . dimensionless

i1 i2 f1 f2

b y y n

FV AA

Angle of attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . degree Input variable 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . m/s Input variable 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . degree Output variable 1 . . . . . . . . . . . . . . . . . . . . . . . dimensionless Output variable 1 . . . . . . . . . . . . . . . . . . . . . . . dimensionless Membership functions . . . . . . . . . . . . . . . . . . dimensionless Position of single tone . . . . . . . . . . . . . . . . . . . variable unit Measured value . . . . . . . . . . . . . . . . . . . . . . . . . . . variable unit Predicted value . . . . . . . . . . . . . . . . . . . . . . . . . . . variable unit Number of observation . . . . . . . . . . . . . . . . . dimensionless Relative error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . percentage Measured mean value . . . . . . . . . . . . . . . . . . . . variable unit Goodness of t . . . . . . . . . . . . . . . . . . . . . . . . . . dimensionless Free stream velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . m/s Angle of attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . degree

drag by about 7%, the Spiroid, on the same aircraft, will reduce drag more than 10%. The advantages of single winglets for small transports were investigated by Robert Jones [11], on which they can provide 10% reduction in induced drag compared with elliptical wings. Winglets are being incorporated into most new transports, including the Gulfstream III and IV business jets [4], the Boeing 747-400 and McDonnell Douglas MD-11 airliners, and the McDonnell Douglas C-17 military transport. The rst industry application of the winglet concept was in sailplane. The Pennsylvania State University (PSU) 94-097 airfoil has been designed for use on winglets of high-performance sailplanes [14]. To validate the design tools, as well as the design itself, the airfoil was tested in the Penn State Low-Speed, LowTurbulence Wind Tunnel from Reynolds numbers of 2.4 105 to 1.0 106 . Performance predictions from two well-known computer codes are compared to the data obtained experimentally, and both are found to generate results that are in good agreement with the wind tunnel measurements. Another investigation was carried out on wing tip airfoils by J.J. Spillman at the Craneld Institute of technology in England [20]. He investigated the use of one to four sails on the wingtip fuel tank of a Paris MS 760 Trainer Aircraft. Experiments on ight test conrmed the wind tunnel tests and demonstrated shorter takeoff rolls and reduced fuel consumption [22]. Spillman later investigated wingtip vortex reduction due to wing tip sails, and found lower vortex energy 400700 m behind the aircraft, although the rate of decay beyond that was somewhat lower [21]. There has been limited investigation of multiple winglets for aircraft. The split-tip design [12] by Heinz Klug for an aircraft wing is considered a primitive multiple winglets which was created to exploit the non-planar wake geometry by reducing induced drag and wing stress. Vance [23] who is a biologist with an aerodynamic background has done extensive investigation of the split wingtips of soaring birds and he has demonstrated that the tip slots of soaring birds reduce induced drag and increase the span factor of the wings. He found remarkable improvements of slotted wingtips compared with conventional wing with a Clark Y airfoil and he investigated that with the same increase in angle of attack, the Clark Y tip increased the base wing drag by 25%, while the feathered tip actually reduced 6% drag. To improve the performance of a wing, the multi-winglet [19] design was evaluated to demonstrate its advanced performance potential over the baseline wing and an equivalent single winglet. The results of their wind tunnel testing show that certain multiwinglet congurations reduced the wing induced drag and im-

proved L / D by 1530% compared with the baseline 0012 wing. In Europe, an extension to the wing tip airfoils has been developed called WING-GRID [18]. WING-GRID is a set of multiple wing extensions added to the wing. These small wings are added at various angles so that their tip vortices do not interact to form a strong vortex. These smaller vortices dissipate the vortex energy so that the lift distribution is modied and the induced drag of the wing is reduced. But this concept is limited, since it is not able to change conguration in ight to optimize drag reduction. Aerodynamic characteristics for the aircraft model with and without winglet having NACA wing No. 65-3-218 has been explained [16]. An interaction matrix method has also been presented to revalidate the calibration matrix data provided by the manufacturer of the six-component external balance. The calibration of free stream velocity and ow quality in the test section has been established and documented [8]. The aircraft model used for the present study consists of a cylindrical body with NACA 65-3-218 airfoil rectangular wing. The aircraft model has a span of 0.66 m and a chord of 0.121 m. The elliptical shaped winglets (Fig. 1) were designed of wood with chord length of 0.121 m, which matches the wing chord length. Like many other real-world optimization, at present, various techniques exist in soft computing method such as statistics, machine learning, neural network, simulated annealing (SA), genetic algorithms, and fuzzy data analysis. Currently articial intelligence methods have largely been used in the different areas including the aircraft industries. In the aircraft era, many expert systems were designed for predicting the aerodynamic characteristics of the aircraft. Based on the studies on characteristics of the aircraft, an intelligent system using Fuzzy Logic was proposed to predict the aerodynamic characteristics of the aircraft model. Fuzzy Logic has been applied successfully to a large number of expert applications. Fuzzy expert system (FES), a relatively new, intelligent, knowledge based technique performs exceptionally well in non-linear, complex systems [1,3]. This work presents the model of fuzzy system, comprising the control rules and term sets of variable with their relates fuzzy sets, in which classical set theory can be extended to handle partial memberships, enabling to express vague human concepts using fuzzy sets and also describe the corresponding inference systems based on fuzzy rules [15]. In fuzzy rule-based systems, knowledge is represented by ifthen rules. The aim of this study was the construction of fuzzy knowledge-based models for the prediction of aerodynamic characteristics of the aircraft model by controlling free stream velocities and angle of attack based on the Mamdani approach. A comparative performance analysis of this

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

597

Fig. 1. Schematic diagram of the aircraft model with winglet at 60 (conguration 3).

Fig. 2. Aircraft model with winglet at 60 (conguration 3) at wind tunnel test section.

Fig. 3. Calibration rig mounted on the oor of the wind tunnel test section.

approach, by sampling data collected from the operation, was used to validate the fuzzy models. 2. Methodology Experiments were conducted in the Aerodynamics Laboratory Faculty of Engineering (University Putra Malaysia) with subsonic wind tunnel of 1000 mm 1000 mm rectangular test section and 2500 mm long. Fig. 2 shows a photograph of the aircraft model with elliptical shaped winglet, which is mounted horizontally in the test section of the wind tunnel. The wind tunnel can be operated at a maximum air speed of 50 m/s and the turntable has a capacity for setting an angle of attack of 14 . During the test the lift and drag forces were measured using the six-component external balance which was calibrated before test carried out. 2.1. Calibration of the balance Calibration of the six-component balance has been done to check the calibration matrix data provided by the manufacturer. Fig. 3 shows a photograph of the calibration rig used for the validation of calibration matrix, which is mounted on the upper platform of the balance in place of model. The relationship between signal readings, L i and the loads, F i applied on the calibration rig are

given by the following matrix equation, the detailed procedure of calibration using MATLAB software is explained elsewhere [16].

{ L i } = [ K i j ]{ F i }

(1)

Where, [ K i j ] is the coecient matrix, { L i } is the signal matrix, and { F i } is the load matrix. For the loads applied step by step at the various locations on the calibration rig (Fig. 3), equivalent forces and moments indicated { F i } were obtained in the X , Y , and Z directions (Table 1). Furthermore, for the equivalent loading, the corresponding sensor readings from all the load cells indicated { L i } were recorded as shown in Table 2. Using the data of loads applied, F i (Table 1) and the signal output, L i (Table 2), six equations were obtained for Step I as follows:

K 12 9.653 K 14 0.981 + K 16 1.638 = 0.1364 K 22 9.653 K 24 0.981 + K 26 1.638 = 12.7883 K 32 9.653 K 34 0.981 + K 36 1.638 = 22.6336 K 42 9.653 K 44 0.981 + K 46 1.638 = 14.5757 K 52 9.653 K 54 0.981 + K 56 1.638 = 14.5816 K 62 9.653 K 64 0.981 + K 66 1.638 = 0.2070

(2) (3) (4) (5) (6) (7)

In this way the 36 equations were obtained using the applied load and observed signal data for Step I through Step VI in terms

598

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

Table 1 Equivalent forces and moments for different load conditions. Step No. Equivalent force (N) FX Step Step Step Step Step Step I II III IV V VI 0 0 19.375 0 0 0 FY FZ 0 0 0 19.375 29.028 19.375 Equivalent moment (Nm) MX MY 0 0 1.962 3.100 1.560 0 MZ 1.638 1.648 0 0 0 0

9.653 29.028
0 0 0 0

0.981 2.943
0 0 0 3.100

Table 2 Sensor readings corresponding to different load conditions. Step No. Sensor readings (V) L1 Step Step Step Step Step Step I II III IV V VI L2 L3 22.6336 2.3613 1.4470 0.2453 0.0961 0.0883 L4 L5 14.5816 43.7212 18.2760 8.3552 5.4740 7.5851 L6

0.1364 1.0173 20.5343 0.0304 0.0922 0.0432

12.7883 31.7972 1.0173 0.2090 0.1324 0.0069

14.5757 43.4475 18.6645 8.6966 5.7428 1.3077

0.2070 0.9869 35.6682 1.7952 13.8949 3.7670

Table 3 Inverse of coecient matrix (calibration matrix).

0.9478 0.0189 0.4313 0.0259 0.7952 0.0009

0.0071 1.0188 2.4996 1.1536 0.0249 0.0469

0.0122 1.0168 2.5095 1.1540 0.0106 0.0466

0.0104 0.0133 0.4038 0.3514 0.1574 0.0016

0.0058 0.0092 2.1178 0.3606 0.1519 0.0005

0.0042 0.0009 1.0787 0.0000 0.2423 0.0016

Table 4 Calibration matrix supplied with six component external balance.

0.9478 0.0189 0.4311 0.0257 0.7949 0.0009

0.0071 1.0186 2.4992 1.1536 0.0253 0.0471

0.0122 1.0166 2.5091 1.1539 0.0108 0.0466

0.0103 0.0133 0.4038 0.3514 0.1576 0.0016

0.0059 0.0092 2.1177 0.3607 0.1518 0.0005

0.0042 0.0009 1.0787 0.0000 0.2420 0.0016

Fig. 4. Effect of air ow and angle of attack on the aerofoil.

Table 5 Percentage of error. 0.0 0.0 0.0 0.8 0.0 0.0 0.0 0.0 0.0 0.0 1 .6 0.4 0.0 0.0 0.0 0.0 1.9 0.0 1 .0 0.0 0.0 0.0 0.1 0.0 1.7 0 .0 0 .0 0 .0 0 .1 0 .0 0.0 0.0 0.0 0.0 0.1 0.0

of 36 unknowns. These unknowns were solved using MATLAB software giving the coecient matrix K i j . The calibration matrix is obtained by nding the inverse of K i j , coecient matrix. The values of the inverse of coecient matrix are given in Table 3. Table 4 shows the data of the calibration matrix supplied with six component external balance software. An excellent matching of the calibration matrix with 2% error (Table 5) has been observed as obtained in the study with the one provided by the supplier with the six component external balance software. 2.2. Theoretical models and analyses The tests were conducted for free-stream velocity of 21.36 m/s, 26.76 m/s, and 32.15 m/s respectively, without and with winglet of three congurations (conguration 1: without winglet, conguration 2: winglet at 0 and conguration 3: winglet at 60 ). The

ambient pressure, temperature and humidity are recorded using barometer, thermometer, and hygrometer respectively for the evaluation of air density, air viscosity and hence Reynolds number for each of free-stream velocity in the laboratory environment by using Eqs. (8)(10). Longitudinal tests were carried out at an angle of attack ranging from zero degree to 14 with an increment of 2 . The angle of attack can be dened as the angle between the chord line of the aerofoil section and the direction of motion relative to the undisturbed uid as shown in Fig. 4. The lift coecient (Table 6) and drag coecient (Table 7) were obtained by using Eqs. (11)(12) from the experimental results as per the procedure published by the author elsewhere [8]. Using equations of state for perfect gas the air density, in kg/m3 is dened as

p RT

(8)

Where, p is the absolute pressure in N/m2 , T is the temperature in K, and R is the gas constant of air in Nm/(kg)(K). Reynolds number based on the chord length is dened

Re =

v c

(9)

Where, v is the free stream velocity in m/s; is the dynamic viscosity in kg/(m)(s) and c is the chord length in m.

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

599

Table 6 Lift coecients from experimental data. S. No. Congurations Reynolds number 105 Lift coecient, C L Initial angle of attack 0 0.237 0.259 0.306 0.299 0.327 0.359 0.386 0.394 0.416 Stall angle of attack 8 0.805 0.817 0.879 0.829 0.889 0.934 0.930 0.934 0.987 Final angle of attack 14 0.657 0.584 0.733 0.641 0.700 0.713 0.729 0.815 0.885

Conguration 1 (without winglet)

1.7 2.1 2.5 1.7 2.1 2.5 1.7 2.1 2.5

Conguration 2 (winglet at 0 )

Conguration 3 (winglet at 60 )

Table 7 Drag coecients from experimental data. S. No. Congurations Reynolds number 105 Drag coecient, C D Initial angle of attack 0 0.085 0.083 0.065 0.053 0.050 0.049 0.070 0.058 0.047 Stall angle of attack 8 0.156 0.149 0.135 0.080 0.079 0.074 0.108 0.089 0.078 Final angle of attack 14 0.249 0.275 0.211 0.136 0.140 0.128 0.166 0.153 0.134

Conguration 1 (without winglet)

1.7 2.1 2.5 1.7 2.1 2.5 1.7 2.1 2.5

Conguration 2 (winglet at 0 )

Conguration 3 (winglet at 60 )

Fig. 5. The basic structure of the fuzzy system.

The air viscosity, is determined using the Sutherlands equation [2] described below

= 1.458 106

T 15 T + 110.4

(10)

Where, T is the temperature in K. Lift coecient is dened as [2]

CL =

L
1 2

v 2 S
D

(11)

Drag coecient is dened as [2]

CD =

1 2

v 2 S

(12)

In Eqs. (11)(12), L is the lift force in N, D is the drag force in N, is the air density in kg/m3 , v is the free stream velocity in m/s, c is the chord length in m, and S is reference area in m2 . 3. Fuzzy expert system There are a number of different techniques that would work here and therefore a design choice must be made. Some of the

techniques require a relatively accurate model of the system in order to develop a satisfactory system. Fuzzy expert system, on the other hand, does not require a model of the system. Instead, they rely on the knowledge of an expert for the particular system. Therefore, with all of this in mind, a Fuzzy Logic expert system is introduced for the prediction of aerodynamic characteristics of the aircraft model. The main advantage of Fuzzy Logic is that it can be tuned and adapted if necessary, thus enhancing the degree of freedom of the system [17,10]. The general conguration of the fuzzy expert system, which is divided into four main parts, is shown in Fig. 5. The rst part is the fuzzication, in this part crisp inputs are transformed to fuzzy values. Fuzzy sets enter the inference engine which maps the input values using normalized membership functions. The fuzzy-logic inference engine deduces the proper control action based on the available rule base. The fuzzy control action is translated to the proper crisp value through the defuzzier using normalized membership functions. For implementation of fuzzy values into the of aircraft model by using Fuzzy expert system (FES), free stream velocity (FV ) and angle of attack (AA) were used as input parameters and lift coecient (C L ) and drag coecient (C D ) were used as output. For fuzzication of these factors the linguistic variables very low (VL), low (L),

600

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

Table 8 Rule base of fuzzy expert system. Rules Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule Rule 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Input variables FV VL VL VL VL VL L L L L L M M M M M H H H H H VH VH VH VH VH AA VL L M H VH VL L M H VH VL L M H VH VL L M H VH VL L M H VH Output variables CL VL L M H M L H VH VH M L M M H L L L M H M L L L L L CD VL L M H vh VL L M H VH L M M H VH L M M H VH L M M H VH

medium (M), high (H), and very high (VH) were used for the inputs and outputs. In this study, the center of gravity (centroid) method for defuzzication was used because these operators assure a linear interpolation of the output between the rules [9,13]. The units of the used factors were: FV (m/s), AA (degree), and C L and C D are dimensionless. With the fuzzy sets dened, it is possible to associate the fuzzy sets in the form of fuzzy rules. For the two inputs and two outputs, a fuzzy associated memory or decision (also called rule) table is developed as shown in Table 8. Total of 25 rules were formed. 3.1. Fuzzication The rst block inside the fuzzy expert system (FES) is fuzzication, which converts each piece of input data to degrees of membership by a lookup in one or several membership functions. The fuzzication block thus matches the input data with the conditions of the rules to determine how well the condition of each rule matches that particular input instance. There is a degree of membership for each linguistic term that applies to that input variable. 3.2. Membership functions Using MATLAB FUZZY Toolbox, prototype triangular fuzzy sets for the fuzzy variables, namely, free stream velocity (FV ), angle of attack (AA), lift coecient (C L ) and drag coecient (C D ) are set up. The term of parameters (membership functions) are presented in Fig. 6(a), (b), (c) and (d). The membership values used for the FES were obtained from the formulas presented analytically below. These membership functions helped in converting numeric variables into linguistic terms. There is a degree of membership for each linguistic term that applies to that input variable. These formulas were determined by using measurement values. The linguistic expressions and membership functions for inputs and output obtained from the developed rules and following formula.

20i 1 VL (i 1 ) = 5 ; 15 i 1 20 i 1 15 ; 15 i 1 20 (i ) = 5 L 1 25i 1 ; 20 i 1 25 5 i 1 20 ; 20 i 1 25 5 M (i 1 ) = 30i FV (i 1 ) = 1 ; 25 i 1 30 5 i 1 25 ; 25 i 1 30 5 H (i 1 ) = 35i 1 ; 30 i 1 35 5 i 1 30 VH (i 1 ) = 5 ; 30 i 1 35 3 .5 i 2 VL (i 2 ) = 3.5 ; 0 i 2 3.5 i2 ; 0 i 2 3.5 L ( i 2 ) = 3 .5 3 .5 i 2 ; 3.5 i 2 7 3 .5 i 2 3 .5 ; 3.5 i 2 7 3 .5 AA (i 2 ) = M (i 2 ) = 10.5i2 ; 7 i 2 10.5 3 .5 i 2 7 ; 7 i 2 10.5 H (i 2 ) = 3 .5 14i 2 ; 10.5 i 2 14 3 .5 i 2 10.5 VH (i 2 ) = 3.5 ; 10.5 i 2 14 VL ( f 1 ) = 0.4 f 1 ; 0.2 f 1 0.4 0 .2 f 1 0 .2 ; 0.2 f 1 0.4 (f )= 0 .2 L 1 0 .6 f 1 ; 0.4 f 1 0.6 0 .2 f 1 0 .4 ; 0.4 f 1 0.6 0 .2 C L ( f 1 ) = M ( f 1 ) = 0 .8 f 1 ; 0.6 f 1 0.8 0 .2 f 1 0 .6 ; 0.6 f 1 0.8 0 .2 H ( f 1) = 1 f 1 ; 0.8 f 1 1.0 0 .2 f 1 0 .8 VH ( f 1 ) = 0.2 ; 0.8 f 1 1.0 VL ( f 2 ) = 0.072 f 2 ; 0.04 f 2 0.072 0.032 f 2 0.04 ; 0.04 f 2 0.072 (f )= 0.032 L 2 0.104 f 2 ; 0.072 f 2 0.104 0.032 f 2 0.072 ; 0.072 f 2 0.104 0.032 C D ( f 2 ) = M ( f 2 ) = 0.136 f 2 ; 0.104 f 2 0.136 0.032 f 2 0.104 ; 0.104 f 2 0.136 0.032 H ( f 2) = 0.168 f 2 ; 0.104 f 2 0.168 0.032 f 2 0.136 VH ( f 2 ) = 0.032 ; 0.136 f 2 0.168

(13)

(14)

(15)

(16)

3.3. Inference step: determining conclusions The determination of conclusion is taken when the rules that are applied to deciding what the output to the plant (aircraft model) should be. To do this, the recommendations of each rule are considered independently. Then later all the recommendations from all the rules are combined to determine the C L and C D inputs to the aircraft model. In defuzzication stage, truth degrees () of the rules were determined for the each rule by aid of the min and then by taking max between working rules. For example, for FV = 27 m/s and AA = 8 , the rules 13, 14, 18 and 19 will be red. To get the fuzzy inputs the values for FV and AA are substituted into Eqs. (13) and (14) and values are obtained as

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

601

(a)

(b)

(c)

(d)

Fig. 6. (a) Prototype membership functions for free stream velocity ( F V ). (b) Prototype membership functions for angle of attack ( A A ). (c) Prototype membership functions for lift coecient (C L ). (d) Prototype membership functions for drag coecient (C D ).

Fig. 7. Control surface of the fuzzy inferring system.

M (FV ) = 0.6, M (AA) = 0.33,

H (FV ) = 0.4 H (AA) = 0.67

The strength (truth values) of the four rules are obtained as

For rule (13) the consequent is lift coecient (C L ) and drag coecient (C D ) are medium. The membership function for the conclusion reached by rule (13), which is denoted as 13 , is given by

13 = min M (FV ), M (AA) = min(0.6, 0.33) = 0.33 14 = min M (FV ), H (AA) = min(0.6, 0.67) = 0.6 18 = min H (FV ), M (AA) = min(0.4, 0.33) = 0.33 19 = min H (FV ), H (AA) = min(0.4, 0.67) = 0.4

13 (C L ) = min 0.33, M (C L ) , and 13 (C D ) = min 0.33, M (C D )

Similarly, the other membership functions for the conclusion reached by rule (14), (18) and (19) can be obtained.

602

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

3.4. Converting decisions into actions: defuzzication module In this stage defuzzication operation is considered that is the nal component of the fuzzy controller. Defuzzication operates on the implied fuzzy sets produced by the inference mechanism and combines their effects to provide the most certain controller crisp output (plant input). Then the one output denoted by C L and C D can be calculated that best represents the conclusions of the fuzzy controller that are represented with the implied fuzzy sets. Due to its popularity, the center of gravity (COG) defuzzication method is used for combing the recommendations represented by the implied fuzzy sets from all the rules [15]. The output membership values are multiplied by their corresponding singleton values and then are divided by the sum of membership values.
crisp

Fig. 8. Lift coecients for conguration 1 (without winglet).

CL

crisp

= CD

crisp

b i i

(17)

Where b i is the position of the singleton in i the universe, and (i ) is equal to the ring strength of truth values of rule i. Using Eq. (17) with membership values obtained from the rules, the lift coecient (C L ) and drag coecient (C D ) could be obtained as 0.73 and 0.109, respectively. In addition, the predictive ability of developed system was investigated according to mathematical and statistical methods. In order to determine the relative error ( ) of system, the following equation was used:
n

=
i =1

y y 100%
y n

(18)

Fig. 9. Lift coecients for conguration 2 (winglet at 0 ).

Where n is the number of observations, y is the measured value, and y is the predicted value. The relative error gives the deviation between the predicted and measured values and it is required to reach zero. In addition, goodness of t () of predicted system was calculated by following equation:

4. Results and discussions Wind-tunnel measurements were done using the aircraft model at Reynolds numbers 1.7 105 , 2.1 105 , and 2.5 105 for corresponding free-stream velocity of 21.36 m/s, 26.76 m/s, and 32.15 m/s respectively. Figs. 813 show the experimental Wind Tunnel Test results obtained for C L and C D versus the angle of attack for three Reynolds numbers and three congurations (conguration 1: without winglet, conguration 2: winglet at 0 and conguration 3: winglet at 60 ). Fig. 8 shows that the lift increases with increase in angle of attack to a maximum value and thereby decreases with further increase in angle of attack for the aircraft model without winglet (conguration 1). The maximum value of the lift coecient appears to increase with the increase in Reynolds number. At the maximum value of the angle of attack the lift coecient characteristic has a mixed behavior e.g. the value of the lift coecient rst decreases with increase in Reynolds number and then increases with further increase in Reynolds number. At the maximum angle of attack of 14 the lift coecients are 0.657, 0.584, and 0.733 respectively for the Reynolds numbers of 1.7 105 , 2.1 105 , and 2.5 105 . The reason for a drop in lift coecient beyond a certain angle of attack e.g. 8 is probably due to the ow separation, which occurs over the wing surface instead of having a streamlined laminar ow there. This condition is called stalling condition and the corresponding angle of attack is called stalling angle. The stalling angle happens to be approximately 8 for all the Reynolds numbers under the present study. The lift coecients data the other two congurations i.e. conguration 2 (winglet at 0 ), and conguration 3 (winglet at 60 ) are given in Figs. 9 and 10 respectively. However, for both congurations 2 and 3, a similar pattern is observed. For the maximum Reynolds number of 2.5 105 the lift coecients for congura-

= 1

n i =1 ( y n i =1 ( y

y )2 y )2

(19)

Where y is the mean of measured values. The goodness of t also gives the ability of the developed system and its highest value is 1.
3.5. Control surface With two inputs and two outputs the inputoutput mapping is a surface. Using MATLAB the fuzzy control surfaces are developed as shown in Figs. 7(a) and 7(b) and these are the mesh plot of an example relationship between free stream velocity (FV ), and angle of attack (AA) on the input side, and controller outputs lift coefcient (C L ) and drag coecient (C D ) on the output side. These control surfaces are the output plotted against the two inputs, and display the range of possible defuzzied values for all possible inputs. The plot results from the interpolation of rule base with twenty ve rules. The plot is used to check the rules and the membership functions and to see if they are appropriate and whether modications are necessary to improve the output. If necessary, the rule base for the fuzzy sets is modied until the output curves are desired. When a satisfactory system is achieved, the fuzzy program is converted to machine language and downloaded into a microprocessor controller. Although the process seems to be long, it actually is relatively easy to do, and it adds intelligence to a machine.

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

603

Fig. 10. Lift coecients for conguration 3 (winglet at 60 ).

Fig. 12. Drag coecients for conguration 2 (winglet at 0 ).

Fig. 11. Drag coecients for conguration 1 (without winglet).

Fig. 13. Drag coecients for conguration 3 (winglet at 60 ).

tion 2 (Fig. 9) and for conguration 3 (Fig. 10) are 0.934 and 1.018 respectively corresponding to an angle of attack of 8 which is stall angle of attack also. The drag coecients of the aircraft model without winglet (conguration 1) under test for all Reynolds numbers are shown in Fig. 11. The drag increases slowly with increase in angle of attack to a certain value and then it increases rapidly with further increase in angle of attack. The value of the drag coecient appears to decrease with the increase in Reynolds number. At the maximum angle of attack of 14 the drag coecients are 0.249, 0.275, and 0.211 respectively for the Reynolds numbers of 1.7 105 , 2.1 105 , and 2.5 105 . The rapid increase in drag coecient, which occurs at higher values of angle of attack, is probably due to the increasing region of separated ow over the wing surface, which creates a large pressure drag. The drag coecients data for the other two congurations i.e. conguration 2 (winglet at 0 ), and conguration 3 (winglet at 60 ) are given in Figs. 12 and 13. It is noticed that for both congurations 2 and 3 a similar pattern has been observed. In general it is observed that the coecient of drag decreases with the increase of Reynolds number. At maximum Reynolds number of 2.5 105 and at 0 angle of attack the drag coecients for the conguration 2 (Fig. 12) and conguration 3 (Fig. 13) are 0.049 and 0.047 respectively. Decisively it can be said that the winglet at 60 (conguration 3) has, overall, the best performance, giving about 6% increase in lift curve slope as compared to other congurations [8]. The results of the developed fuzzy expert system (FES) were compared with the experimental results for the conguration 3 (Tables 910 and Figs. 1417). For lift coecient, the mean of measured and predicted values were 0.751 and 0.712, respectively. For

drag coecient, the mean of measured and predicted values were 0.092 and 0.093, respectively. The correlations between measured and predicted values (from FES model) of lift coecient and drag coecient in different angle of attack were given in Figs. 1417. The relationships were signicant for all parameters. The correlation coecient of relationships were 0.982 for lift coecient and 0.985 for drag coecient. The mean relative error of measured and predicted values (from FES model) were 6.52% for lift coecient and 4.74% for drag coecient. For all parameters, the relative error of predicted values was found to be less than the acceptable limits (10%) [3]. The goodness of t of prediction (from FES model) values were found as 0.94 for lift coecient and 0.98 for drag coecient. For all parameters the goodness of t were close to 1.0 as expected. Based on the discussions the above indices indicate that the system is qualied to replace the work of an operator by using intelligent control system. Since the fuzzy expert system is found to be valid as compared to experimental data, hence the fuzzy program can be converted to machine language (or other real time code) and downloaded into a microprocessor controller. The microprocessor then runs the machine or the system based on the fuzzy intelligent program. Thus, an intelligent control system can be achieved by using fuzzy logic controller (FLC) which will operate the system instead of an operator [6]. Furthermore, this investigation provides a better understanding for the winglet concept and its inclusion to the wing of aircraft and motivates for using the fuzzy expert system for predicting aerodynamic characteristics. The important benet for using the fuzzy expert system lies in the opportunity for aircraft control systems to handle the parameters of

604

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

Table 9 Lift coecients data from FES for conguration 3 (winglet at 60 ). S. No. Sample data collection Reynolds number 105 Lift coecient, C L Initial angle of attack 0 0.402 0.405 0.432 0.386 0.394 0.416 Stall angle of attack 8 0.890 0.898 0.979 0.930 0.934 0.987 Final angle of attack 14 0.700 0.735 0.835 0.729 0.815 0.885

Predicted data (from FES)

1.7 2.1 2.5 1.7 2.1 2.5

Measured data (from experiment)

Table 10 Drag coecients data from FES for conguration 3 (winglet at 60 ). S. No. Sample data collection Reynolds number 105 Drag coecient, C D Initial angle of attack 0 0.065 0.060 0.051 0.070 0.058 0.047 Stall angle of attack 8 0.102 0.094 0.085 0.108 0.089 0.078 Final angle of attack 14 0.143 0.148 0.123 0.166 0.153 0.134

Predicted data (from FES)

1.7 2.1 2.5 1.7 2.1 2.5

Measured data (from experiment)

Fig. 14. The measured and predicted values of lift coecient vs angle of attack for conguration 3.

Fig. 16. Correlation between measured and predicted values of lift coecient for conguration 3.

Fig. 15. The measured and predicted values of drag coecient vs angle of attack for conguration 3.

Fig. 17. Correlation between measured and predicted values of drag coecient for conguration 3.

A. Hossain et al. / Aerospace Science and Technology 15 (2011) 595605

605

resistance and tyresurface friction taking into account changeable external conditions which are not directly related to an aircraft. 5. Conclusion In this paper, a robust fuzzy knowledge based rules has been developed to predict aerodynamic characteristics of an aircraft model with and without winglet and nally it was compared with the experimental data. Both experimental and predicted results were found to be valid within the acceptable limits. In addition, calibration of the wind tunnel was performed and was validated with the software used for the experiments. Further progress in the controller is, in the development of controller for load cell monitoring that could reect more characteristics of the aircraft model. However, the conclusions drawn from this investigation are as follows: (a) Mean relative errors were found as 6.52% for lift coecient and 4.74% for drag coecient which were less than the acceptable limit of 10%. (b) Goodness of t was found as 0.94 for lift coecient and 0.98 for drag coecient which were close to 1.0 as expected. (c) The developed model can be used as a reference for further aerodynamic characteristics studies on aircraft model. The lift coecient and drag coecient prediction from the system shows the good performance of the developed FES model and hence warrant the novelty of this work. References
[1] A. Al-Anbuky, S. Bataineh, S. Al-Aqtash, Power demand prediction using fuzzy logic, Control Engineering Practice 3 (9) (1995) 12911298. [2] John J. Bertin, Aerodynamics for Engineers, Prentice Hall Inc., New Jersey, 2002. [3] K. Carman, Prediction of soil compaction under pneumatic tires a using fuzzy logic approach, Journal of Terramechanics 45 (2008) 103108. [4] Reuben M. Chandrasekharan, William R. Murphy, Frank P. Taverna, Charles W. Boppe, Computational aerodynamic design of the Gulfstream IV wing, AIAA85-0427, 1985. [5] Reginald V. French, Vortex reducing wing tip, U.S. Patent 4,108,403, 1978. [6] M. Gopal, Digital Control and Sate Variable Methods: Conventional and Intelligent Control Systems, 3rd edition, Tata McGrawHill Education Pvt. Ltd., 2009.

[7] Louis B. Gratzer, Spiroid-tipped wing, U.S. Patent 5,102,068, 1992. [8] A. Hossain, P.R. Arora, A. Rahman, A.A. Jaafar, A.K.M.P. Iqbal, Analysis of aerodynamic characteristics of an aircraft model with and without winglet, Jordan Journal of Mechanical and Industrial Engineering 2 (3) (2008) 143150. [9] A. Hossain, A. Rahman, M. Rahman, S.K. Hassan, J. Hossen, Prediction of power generation of small scale vertical axis wind turbine using fuzzy logic, Journal of Urban and Environmental Engineering 3 (2) (2009) 4351. [10] V. Ivanov, B. Shyrokau, K. Augsburg, V. Algin, Fuzzy evaluation of tyresurface interaction parameters, Journal of Terramechanics 47 (2010) 1131130. [11] Robert T. Jones, Improving the eciency of smaller transport aircraft, in: 14th Congress of the International Council of the Aeronautical Sciences, Proceedings, vol. 1, Toulouse, France, 1984. [12] Heinz G. Klug, Auxiliary wing tips for an aircraft, U.S. Patent 4722499, February 1988. [13] T. Marakoglu, K. Carman, Fuzzy knowledge-based model for prediction of soil loosening and draft eciency in tillage, Journal of Terramechanics 47 (2010) 173178. [14] M.D. Maughmer, T.J. Swan, S.M. Willits, The design and testing of a winglet airfoil for low-speed aircraft, AIAA Paper 2001-2478, 2001. [15] Kevin M. Passino, Stephen Yurkovich, Fuzzy Control, AddisonWesley Longman, Inc., Menlo Park, CA, USA, 1998. [16] R.A. Prithvi, A. Hossain, A.A. Jaafar, P. Edi, T.S. Younis, M. Saleem, Drag reduction in aircraft model using elliptical winglet, Journal of IEM, Malaysia 66 (4) (2005) 18. [17] A. Rajagopalan, G. Washington, G. Rizzani, Y. Guezennec, Development of fuzzy logic and neural network control and advanced emissions modeling for parallel hybrid vehicles, Center for Automotive Research, Intelligent Structures and Systems Laboratory, Ohio State University, USA, December 2003. [18] La.U. Roche, S. Palffy, WING-GRID, a novel device for reduction of induced drag on wings, in: Proceedings of ICAS 96, Sorrento, September 1996. [19] M.J. Smith, N. Komerath, R. Ames, O. Wong, J. Pearson, Performance analysis of a wing with multiple winglets, AIAA Paper-2001-2407, 2001. [20] J.J. Spillman, The use of wing tip sails to reduce vortex drag, Aeronautical Journal (September 1978) 387395. [21] J.J. Spillman, M.J. Fell, The effects of wing tip devices on (a) the performance of the Bae Jetstream (b) the far-eld wake of a Paris Aircraft, Paper 31A, AGARD CP No. 342, Aerodynamics of Vortical Type Flows in Three Dimensions, April 1983, pp. 31A-111. [22] J.J. Spillman, H.Y. Ratcliffe, A. McVitie, Flight experiments to evaluate the effect of wing-tip sails on fuel consumption and handling characteristics, Aeronautical Journal (July 1979) 279281. [23] A.T. Vance, Gliding birds: Reduction of induced drag by wing tip slots between the primary feathers, Journal of Experimental Biology 180 (1) (1993) 285310. [24] R.T. Whitcomb, A design approach and selected wind-tunnel results at high subsonic speeds for wing-tip mounted winglets, NASA TN D-8260, 1976. [25] R.T. Whitcomb, Methods for reducing aerodynamic drag, NASA Conference Publication 2211, in: Proceedings of Dryden Symposium, Edwards, CA, 1981. [26] J.E. Yates, C. Donaldson, Fundamental study of drag and an assessment of conventional drag-due-to-lift reduction devices, NASA Contract Rep. 4004, 1986.